Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Zebec x Norbert Network AMA
Recorded: Oct. 31, 2023 Duration: 0:34:33Space Recording
Full Transcription
Norbert, Ross, you'd like to give a quick intro on yourself?
Oh, okay.
You know, you know, you know,
you're in a lot of people, you know, you're in a lot of people, you know, you're in the
for, you know, you're in a mal form, you know, you're in the
But from a customer standpoint, really data privacy is almost even more important, right?
So Norbert never sells customer data or uses it to enhance any of its own internal AI models or applications.
One in every three generative AI platforms take customer data to enhance their own models, even if they don't directly sell customer data.
And some of these platforms even will have clauses in their agreements that say they, quote, reserve the right to sell or use customer data at some point down the line.
You may see terms marketed such as synthetic AI or AI powered models.
You know, this is just an example of them even putting right on their website sort of with suspicious marketing terms that they are indeed taking past customer projects
and incorporating that data into their own models.
I'm not sure if anyone on the call is aware of this, but in the U.S., we've had two cases here where a couple of IBM engineers got fired
and then subsequently sued because they were using ChatGPT to try to help them write source code,
accidentally uploaded a bit of IBM's own source code, which then made that publicly available through that backdoor.
There have also been numerous cases of law firms and attorneys using ChatGPT and other generative AI solutions to write legal briefs and documents,
accidentally exposing their clients' confidential information to everyone else that ends up using it.
So obviously, whether it's customer-facing applications such as e-commerce platforms or CRMs, oh, I'm sorry.
We provide full end-to-end encryption of any customer data in transit both to and from Norbert.
And we never keep customer data for any reason at the conclusion of our projects, which is indeed included in our SLA and contractual agreements.
That's awesome. No, that's phenomenal and actually kind of crazy, the example you gave out.
So the fact that, you know, that Norbert never sells customer data or uses it, you know, to kind of enhance its own models or AI apps and whatnot,
I think it's phenomenal. I mean, yeah.
What kinds of data can Norbert handle?
Great question. So really, we can handle any type of data set for labeling or annotation.
And really, there's only a couple categories when you really think about it.
You know, most common is text, which we see in the chat GPT and which most people are probably used to and have been exposed to.
But we can handle static images. We can handle static video.
But we can even handle 3D imaging or AR and VR data for labeling and annotation.
I mentioned this during my intro, but our founding team has both ownership and executive experience in the fintech, blockchain, HR, business intelligence,
several other niche industries, enabling us to provide guidance and expertise in a bunch of different AI and data labeling projects.
So whether it is customer facing applications such as e-commerce platforms or CRMs or internal use cases, shipping and logistics, business analytics,
we have the capability to handle any AI project in virtually any industry or use case.
That's what's up. No, that's super cool.
And yeah, let's kind of, I kind of want to touch a little bit on, wait, actually, I got a question right here from one of our team members.
Can Norbert handle like chem formulas, mind the chance?
Absolutely.
There you go. Anything, anything, Elena.
So, yeah. And what's, I kind of want to touch on, on what is reinforcement learning from human feedback?
Sure. So this is a term that's become popular, well, at least from a marketing standpoint with us and a lot of our competitors,
as far as just letting a little, shedding a little bit more light into a sort of new popular buzzword that's caught the industry by storm.
So RLHF, which was mentioned already is reinforcement learning from human feedback.
So the actual formal definition is it's a machine learning technique that involves training reinforcement learning agents with the help of human feedback or guidance.
In traditional reinforcement learning, agents are learned by trial and error, exploring various actions and receiving rewards based on their actions in the environment.
And RLHF incorporates human feedback to expedite and improve the learning process.
So really, in a nutshell, it's trying to break the AI to make the AI better.
You know, we talked about the example with Pantene Pro-V, but really what it is, is we have an algorithm that's being used to search for or produce data or results on X.
Well, the initial input made X not very accurate or not very equal to what we were searching for.
So therefore, we have humans come in.
A really good example is you're using ChatGPT.
Maybe this is a poor example, but let's say you're a high school student using ChatGPT to write an English paper for you.
And you give it your input and you are not satisfied with the result.
It didn't touch on the subject or topic very well at all.
So we would have humans go in, analyze the data set that was initially used to input to produce those results, and then we would either label it, annotate it, maybe even try to break it by asking the question in a different way, and basically just fine-tuning that language learning model or AI algorithm to make sure that it does produce more specific, finite, and accurate results based on the user input.
That's what's up.
No, and I think that's a great example.
I'm sure people have used it for that.
What about IP implications?
Does Norber have any or anything?
Well, so really what we do is – so our customers will submit us a request or a project, right?
And generally, the scope or parameters of that project are outlined by the customer.
However, we will obviously take it up front, do sort of an up-front consulting session with our customer, and if we think that the parameters are either too wide for what they want labeled or how they want it labeled,
or in some cases if they might accidentally be trying to expose some IP or black box info themselves, we'll notify the customer of that.
We might even tell them that they need to modify or change their data set up or simply warn them as a precaution because sometimes the customer might want to publish some IP through the backboard as part of their own process.
But either way, we have a team of highly trained experts, data engineers, et cetera, that can kind of take on any project and just give their own two cents as far as what the customer might want to do or should do,
and then just let them take that feedback and let us know what they want us to do.
There you go.
Okay, that's what's up.
I guess this is a question of me, and yeah, it wasn't sent on the doc, but what are some of the plans that you guys have for the rest of the year?
I mean, obviously, we've got like two months left, but even going into 2024, what are some of the things that you guys are working on and want to accomplish?
Great question.
So I myself mentioned that I'm a veteran of the fintech and financial services space.
I recently came back from Money 2020 in Las Vegas, which is our financial services conference here in the U.S.
And so in speaking to a lot of folks, we generated a ton of interest.
And not only are folks interested in just having their data sets labeled, but a lot of them expressed interest in having us potentially build models for them
or do some more fine-tuning and tweaking of their existing models beyond simply just data labeling.
So we launched the company with data labeling as our initial product slash service because it's low-hanging fruit for us.
It's something we can basically turn around in three weeks to a month
and something that we were able to easily implement and roll out to our existing portfolio companies.
However, we've generated so much interest in the fintech compliance KYC space as far as would you guys be able to potentially, you know,
build or tweak a model for KYC or compliance, catch bad actors, catch money launderers, catch scammers.
So we're actively exploring that as we speak.
And potentially by the end of the year, early next year, we'll be able to expand our service offerings beyond simply just data labeling
to handle any type of AI project, whether it's being built from scratch or if it's taking a model
and maybe doing some actual data engineering and tweaking on it.
That's what's up.
By the way, I'm in Vegas, so if you're ever here in Money 2020 next year, definitely let me know.
And I'm in San Diego, so it's a short 45-minute hop, skip, and a flip.
Man, we will talk behind the scenes.
I'll talk to Park, man.
No worries.
But, no, it's been – and how long – when did you join, man?
When did you join?
So me and Sam have been close for roughly three or four years.
We were actually co-investors on another project that I was running sales for.
And as soon as that took off the ground, hit our growth metrics, and got some traction,
was simply looking for a new challenge and a new company to help grow and scale up.
Me and Sam were always in constant contact.
The second he was ready to kind of launch Norbert and do the initial service offering,
we had a conversation.
And here I am talking to you lovely folks today.
There you go.
There you go.
No, Ross, it's been an absolute pleasure to, one, get to know you, your story, and everything.
Did you go to Wisconsin?
Is that you on LinkedIn?
No, that's not me.
That's not you?
I've been to Wisconsin, though.
It's a beautiful state and a beautiful place to both visit and hang out.
There you go.
I'll talk to a part, see if we can set something up.
We'd love to.
Yeah, and how do people get in touch with you guys?
If anybody has, like, any sort of, you know, AI project they want to work on,
especially when you guys are past, like, the data labeling phase,
where can they hit you guys up?
Twitter, any shout-outs?
Well, yeah.
Yeah, so, obviously, you have our Twitter handle, Norbert AI.
Our website is www.norbert.network.
We actually do own the domain for Norbert.ai.
We just decided that since we have this network of in-house data labelers
that we would simply start there for marketing purposes.
But we should have that .ai domain incorporated within the next couple of days.
Everyone can feel free to reach out to me directly, too.
My email is simply ross, R-O-S-S, at norbert.network.
Feel free to shoot me an email at any point, at any time, with any questions,
comments about what we do.
Or, for sure, if you have an AI data labeling project,
would definitely love to get that as well.
But feel free to reach out to me on Twitter here, at Norbert AI.
We have plenty of methods for you to contact us and me personally
directly through the website.
But for anyone listening out there, and let me see.
Can I drop this in a comment right here?
You can.
I think you can.
But if not, what I can do is we can have Parth,
because I don't have access.
I mean, we're just a partner.
So, yeah, have Parth kind of put it on.
A hundred percent.
Well, if anyone's listening, it's simple.
Ross at norbert.network.
Reach out to me anytime, 24-7.
There you guys go.
I think we have a question here.
Let me see.
And then we'll.
I know we got five minutes, so this will probably be the last.
Just see.
All right.
Well, he.
All right.
Internet money.
Are you.
Are you there?
Hello, Internet money.
All right.
Well.
I guess he is not listening.
All right, Ross.
It's been a pleasure talking to you.
Hey, man.
It was great.
Appreciate you with the questions.
Love this format.
Full disclosure.
This is the first AMA I've ever done,
and I certainly don't think it'll be my last.
This was a blast.
No, absolutely.
These are.
These are good,
especially once you start building community on Norbert.
And more people, obviously, like as AI gets bigger,
you guys are going to be a big player in the game.
So AMAs are just good for, you know,
informing people, talking to them.
And you did a great job explaining, like, you know,
difficult concepts, quote, unquote, very easily.
I think that's kind of like the name of the game when it comes to AMAs,
especially in like tough kind of like, you know,
just industries, I guess, like the ones we're in, you know.
My guy.
All right, guys.
Thank you so much for the AMA.
Really appreciate the time, Ross.
And we'll talk soon.
Catch you guys.
Appreciate you, brother.
Yes, sir.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
Bye-bye.
